Welcome to the Machine Intelligence Group!
We want to study a fundamental aspect of artificial intelligence: how to learn to make decisions and to act. Action is the only way that an intelligent agent can change or manipulate its environment in order to solve its tasks. It is fascinating for us to study and develop a general-purpose learning framework that requires minimal human data/knowledge and performs a wide range of tasks.
We are interested in many critical topics and questions raised from the development of such a general-purpose learning framework, including representation learning, world-model learning and reasoning, policy learning and planning, and multi-agent learning. Specifically, we want to build an autonomous agent that can extract relevant information from its perception, learn and reason a world model generalized over tasks, predict the future and consequences of its actions, imagine and generate goals that it could achieve, plan or learn to generate actions to achieve these goals, collaborate and communicate with other agents or humans for performing complex tasks, generalize and share its learning and knowledge with other agents.
To approach these topics, current research in our group is building novel efficient models and methods of deep learning, reinforcement learning, and multi-agent systems, with applications in robotics and video games, drawing insights from other fields, including cognitive science, game theory, information theory, social science, and systems science.